This study conducted an empirical Bayes before-after evaluation of the conversion of at-grade intersections to diamond interchanges. The data included 20 intersections that were converted in Minnesota and North Carolina. Before conversion, 6 of the intersections in Minnesota were stop-controlled, and 4 were signalized. In North Carolina, all the 10 intersections were stop controlled before conversion. The study included reference sites in both Minnesota and North Carolina. The combined results from the two States including all 20 sites indicated that injury and fatal crashes decreased by about 30 percent, PDO crashes increased by about 11 percent, and total crashes decreased by about 8 percent. For the 16 sites that were stop-controlled before conversion and had stop-controlled ramp terminals after conversion, injury and fatal crashes decreased by about 12 percent, PDO crashes increased by about 156 percent, and total crashes increased by about 60 percent. The 4 sites (all from Minnesota) that were signalized before conversion experienced significant reductions in crashes, but the sample size is probably not sufficient to provide a reliable finding for this group.